CN1264996A - Method for pre-processing wireless measured data in cellular mobile communication system - Google Patents

Abstract

A method for pre-processing the wireless measured data in cellular mobile communication system features tha the length-unlimited low-pass filter is used to smooth the measured data and the jump between two continuous data is detected to obtain needed information. A method integrating filter with pattern recognition is disclosed to easily limit the shadow fading to lower than 0.5.

Description

The preprocess method of wireless measured data in a kind of cell mobile communication systems

The invention belongs to wireless communication technology field, is the preprocess method of wireless measured data in a kind of cell mobile communication systems.

In cell mobile communication systems, base station and travelling carriage communicate by letter carry out in, carry out the measurement of radio characteristics to the uplink downlink of Serving cell and the broadcast channel of adjacent cell.The result who measures will be as the foundation of power control and switch decision.

Consider two base stations in the canonical system: base station 1 (BTS-1) and base station 2 (BTS-2).1 to base station 2 moving linearlies, the signal strength signal intensity of its communication as shown in Figure 1 from the base station for travelling carriage.The signal strength signal intensity of travelling carriage continuous base stations detected 1 in motion process and the signal strength signal intensity of base station 2.When the signal strength signal intensity of base station 1 was enough strong or enough weak, system may consider to carry out power control.And when the signal strength signal intensity of base station 2 during greater than the signal strength signal intensity of base station 1 (A point), system may consider to switch.The typical switching is to carry out handover under the situation of the signal strength signal intensity strong a lot (promptly greater than hysteresis margin h, C point among the figure) of 2 signal intensity ratio base station 1 in the base station.Can prevent like this because the travelling carriage that signal fluctuation causes repeating back and forth between two base stations switched-i.e. " table tennis switches ".

Because mobile channel not only is subjected to the influence of path loss, also is subjected to the influence of shadow fading and multipath fading.Because the cycle of measuring generally is far longer than the cycle that multipath fading changes, the result of measurement can think the influence that does not comprise multipath fading.But the cycle general and that slow fading changes in the cycle of measuring is suitable, and therefore directly measurement result can not need be done certain processing directly as the foundation of control.Because the information that the variation of slow fading and path loss overlaps in together is disorderly, therefore cause a large amount of erroneous judgements of system: control that can not triggering system when pressing for as inappropriate the time, carrying out power control or handover.Especially the variation of path loss links together with incident, and it makes system powerless to some incident, and in the presence of the time when very frequent, it is more abominable that the characteristic of system just becomes.

The current actual Measurement and Data Processing method of using mainly contains two kinds: a kind of is the average method of sliding window, and another kind is that average method (Trends in Handover Design IEEECommunication Magazine 1996.5) is asked in the index weighting.They all can be expressed as the form of transversal filter: $X\left(n\right)=\frac{1}{C}\underset{m=0}{\overset{N-1}{\mathrm{\&Sigma;}}}x(n-m)h\left(m\right)---\left(1\right)$ X (n) is the initial data before handling, and X (n) is the data after handling, and C is a normalization coefficient, and N is a slip sampling window length.The sliding window averaging method is by the most extensive employing, its h (m)=1, C=N.The amplitude of its system function is: $\left|H\left(\mathrm{\&omega;}\right)\right|=\left|\underset{m=0}{\overset{m=N-1}{\mathrm{\&Sigma;}}}h\left(m\right)\exp (-\mathrm{jm\&omega;})\right|=\left|\frac{\sin \left(\frac{\mathrm{<figure-callout\; id="2"\; label="N"\; filenames="A0011178000092.png"\; state="\{\{state\}\}">N</figure-callout>}}{2}\mathrm{\&omega;}\right)}{\sin \left(\frac{1}{2}\mathrm{\&omega;}\right)}\right|---\left(2\right)$ The frequency spectrum of its system function such as Fig. 2.In general, this is a kind of simple but very effective data smoothing mode.For this noise like of shadow fading, the speed of travelling carriage is big more, and being correlated with of sampled point is more little, and the broadening of power spectrum is big more, and effect is good more.For static travelling carriage, then not effect in theory.The level and smooth effect of the average method of sliding window also increases along with the number of average sampling point and better.For example for the travelling carriage (this is the very high travelling carriage of this type of sub-district medium velocity) of speed 40 metre per second (m/s)s in the medium sub-district, the gross power of shadow fading was not σ when measurement data was not done any processing ², calculating shows: after carrying out the average method of two sampling points, the gross power of shadow fading is reduced to 3/4ths σ ²Carry out the mean time of three sampling points, the gross power of shadow fading becomes After carrying out four sampling points average, the gross power of shadow fading is Above data show that also average number surpasses after four, to the inhibitory action of shadow fading will be not can be than four sampling points too many improvement on average arranged.Because the increase of average number of samples causes the increase of processing delay, increase simply therefore that average length of window loses more than gain often, the inhibition of shadow fading also is difficult to surpass 1/2nd σ ²

The another kind of method that also compares nature relatively more commonly used is the index weighted-average method.Its h (m)=exp (α m), normalization coefficient is

, the amplitude of system function is: $\left|H\left(\mathrm{\&omega;}\right)\right|=\left|\underset{m=0}{\overset{m=N-1}{\mathrm{\&Sigma;}}}\exp (-\mathrm{\&alpha;m}-\mathrm{jm\&omega;})\right|=\left|\frac{1-\exp [-(\mathrm{\&alpha;}+\mathrm{j\&omega;})N]}{1-\exp (-\mathrm{\&alpha;}-\mathrm{j\&omega;})}\right|---\left(3\right)$ Its frequency response as shown in Figure 3.Average number of samples N among the figure equals 4, and attenuation factor equals 0.5.

Can analyze on scheming, two kinds of methods all are to realize the level and smooth estimation of measurement data by the mode of leaching low frequency, change the width of passband to adjust the elimination effect to noise by adjusting average sampling point number N.In addition, equally by adjusting the time-delay that N changes processing.In the method for exponent-weighted average, it is more that radio-frequency component keeps, because it respects last measurement data sample value more, can improve the problem of handover delay like this, its time delay no longer can equal simply average number of samples purpose half, but the fluctuation of data has increased.

A total shortcoming of transversal filter is its limited length of window.Sampling window length is all smaller in the real system, because limited length of window, the sampling frequency spectrum of path loss and shadow fading all further overlaps, and is unfavorable for that more the separation of frequency spectrum detects.

The objective of the invention is to propose a kind of preprocess method about wireless measured data that in cellular communication system, can improve path loss and shadow fading is suppressed.

From aforementioned analysis as can be known, the unification of data smoothing and reflection data variation is being pursued in the processing of measurement data always.Though existing method is recognized contradiction between the two, do not recognize that both are separable on incident, so all selected trading off between the two.Therefore will study this problem, it is level and smooth at first to want clear and definite what data to need, and what data needs reflection.

In the data sequence of a reality, if wherein have only data that bigger fluctuation is arranged, these data should be smoothedly to fall so.Its cycle of changing of this situation is 2, ignores the shape of wave form varies and in the enterprising line period continuation of time domain, it is a shock pulse that is positioned at 0.5HZ frequency place in power spectrum signal.The power spectrum medium frequency is variation faster greater than the representative of the part of 0.5HZ, therefore can think the power spectrum medium frequency greater than the part of 0.5HZ all be need smoothed.Consider the shape of fluctuation, the fourier expansion that one-period equals the signal of N must comprise the component of cycle smaller or equal to N.If filtering makes signal not comprise the carrier wave of cycle less than N, then length period of waves of signal is affirmed more than or equal to N.Therefore, make period of waves, only need harmonic wave elimination all periods of waves less than N greater than N.

Need on earth in the Measurement and Data Processing level and smooth continuously what data fluctuations, to such an extent as to be can stand channel degradation how long and not cause the communication quality call drop that declines to a great extent according to communicating pair in the communication.Channel degradation can be reflected as the raising of the incorrect probability of acceptance of block of information.For the sampling point in arbitrary moment in the communication process, may be in the fluctuation time width and equal in the fluctuation of n.Because data smoothing, the fluctuation of supposing this width is smoothedly to fall, so communicating pair do not carry out corresponding power control and hand-off process, has caused the deterioration of communication quality thus.Ideally, power control and switching controls can avoid all cycles to worsen greater than 2 the communication quality that fluctuation caused, and system can only handle the fluctuation of width greater than N of fluctuating after the smoothing processing.Obviously the raising of N value will cause the rising of cutting off rate.The choosing of N value depended near the wireless coverage situation the switching point, and under wireless coverage situation preferably, the call drop that causes is negligible in total cutting off rate.

In actual treatment, the frequency spectrum crossover of transversal filter is more serious.Shown in Fig. 2 and 3, kept a large amount of radio-frequency components in the filter system function, make inhibition poor effect to the shadow fading of high frequency, generally be not less than the level of 1/2nd shadow fading power.And adopt the endless low pass filter is better choice, with the cycle less than the fluctuation of N all filtering go out, to reach level and smooth purpose.Compared with the average treatment method, its complexity does not have to improve substantially, but far better to the filtration result of shadow fading, and the gross power of shadow fading can be less than the level of 1/2nd former power very like a cork after the filtering.

For traditional cellular communication, wireless propagation environment changes not quite in cellular coverage, not significantly sudden change, and the variation of path loss signal strength signal intensity is milder, and a spot of delay and error almost do not have influence to conversation.But in Microcell, the turning effect in street is just very outstanding.After travelling carriage turned round through one, in the distance of 20 meters of less thaies, the change of path loss can reach about 15dBm.Such data must be reflected, and delay should be as much as possible little.If this situation can not obtain reflection or be delayed, then cause call drop probably.

Review in the shadow fading, the probability that the quick saltus step situation significantly of signal level intensity takes place then is very little.Shadow fading only stays very smooth low frequency part through after the low-pass filtering, ignores the frequency spectrum crossover that causes owing to periodic sampling, can be approximately the narrowband Gaussian white noise, and its standard variance is: ${\mathrm{\&sigma;}}^{\&prime;2}={\mathrm{\&sigma;}}^2\frac{{\&Integral;}_0^{1/N}\mathrm{\&Psi;}\left(\mathrm{\&omega;}\right)\mathrm{d\&omega;}}{{\&Integral;}_0^{+\&infin;}\mathrm{\&Psi;}\left(\mathrm{\&omega;}\right)\mathrm{d\&omega;}}---\left(4\right)$ A signal strength signal intensity whereabouts saltus step is greater than the Probability p of Δ _△For: $p_{\mathrm{\&Delta;}}={\&Integral;}_{-\&infin;}^{+\&infin;}(\frac{1}{\sqrt{2\mathrm{\&pi;}}{\mathrm{\&sigma;}}^{\&prime;}}{e}^{\frac{-{\mathrm{<figure-callout\; id="2"\; label="s"\; filenames="A0011178000092.png"\; state="\{\{state\}\}">s</figure-callout>}}^2}{2{\mathrm{\&sigma;}}^{\&prime;2}}}\&CenterDot;{\&Integral;}_{-\&infin;}^{s-\mathrm{\&Delta;}}\frac{1}{\sqrt{2\mathrm{\&pi;}}{\mathrm{\&sigma;}}^{\&prime;}}{e}^{\frac{-{\mathrm{<figure-callout\; id="2"\; label="x"\; filenames="A0011178000092.png"\; state="\{\{state\}\}">x</figure-callout>}}^2}{2{\mathrm{\&sigma;}}^{\&prime;2}}}\mathrm{dx})\mathrm{ds}---\left(5\right)$ The whereabouts saltus step of double signal strength signal intensity greater than the probability of Δ can and the like, Δ is that thresholding is handled in the signal level saltus step, general desirable 4-6 is specifically according to system requirements with measure and determine.When N gets 10, when the speed of travelling carriage was got 20 metre per second (m/s)s, the shadow fading gross power was reduced to 1/4th of former power, p after the low-pass filtering _ΔWith the variation of Δ as shown in Figure 4.When Δ=5, p _ΔValue be 0.03.The probability that makes twice saltus step Δ 〉=5 is then certainly less than 10 ^-5Therefore as long as in algorithm, detect quick saltus step, can think that variation by the path loss of motion travelling carriage brings, rather than shadow fading brings, without the level and smooth received signal of directly recovering, just can avoid the delay of smoothing processing, thereby avoid the call drop that causes because of turning effect or other reasons quick saltus step.If only detect a saltus step, data processing section does not bring delay; Detect twice saltus step, then maximum delay is 1; It is nonsensical substantially to detect twice above saltus step.In addition, the increase of 3 percent shadow fading power is seldom for the shadow fading gross power of also not removing after level and smooth: because level and smooth for N=10, shadow fading after level and smooth still has 1 15 of former power, so the increase of the power of 3 percent probability is also little to the influence of smoothness properties.And this increase often highly significant, and pernicious shadow fading changes because it has reflected there and then, may also be necessary if carried out a handover this moment.

In fact the processing of signal level ionization meter data smoothing can be understood as a kind of knowledge representation process: for each data of receiving, according to past and current information, need to judge whether smoothly, how much smoothly need.This judgement is to be based upon signal in the signal strength measurement data and noise separately on the understanding of characteristic, are known more, and it is accurate more to judge.In front, mainly by the analyzed and researched characteristics of signal of correlation between frequency domain and the shadow fading data.Therefore, the present invention proposes, and obtains needed information by the low pass processing of frequency domain and the transition detection between continuous two data.The two independently works separately, can separate fully, so the contradiction between smoothing processing and the processing delay can be separated because of the difference of incident itself.

The preprocess method of the wireless measured data that the present invention proposes, adopt the endless low pass filter exactly, elimination period of waves is less than the fluctuation of signal period N, wireless measured data is carried out smoothing processing: then to continuous two data saltus step detect, data after level and smooth still have the saltus step greater than Δ, then without level and smooth directly recovery received signal.We claim the method for this Measurement and Data Processing method for the identification of filtering binding pattern.Use this method, the inhibition of shadow fading is easy to just can reach surpasses 1/2nd level, and the quick saltus step of signal strength signal intensity can almost not had detecting of time delay.Because the value of N has been expressed the level and smooth degree of data more intuitively, the adjustment of level and smooth degree again even can be continuous, and usual way to give the leeway of this adjustment be seldom several discrete points, therefore also have adjustment meticulous advantage directly perceived more.

The low pass filter that the present invention adopts can be a second order Butterworth wave digital lowpass filter.

The shortcoming that adopts low circulation method is a relatively difficulty of normalization, and its estimation is that inclined to one side estimation is arranged.But it is acceptable that inclined to one side estimation is arranged.Because low frequency part has comprised the power of the signal overwhelming majority, because the power of low-pass filtering loss also will (be pressed continuous spectrum calculating less than one of percentage when N=20, and the power calculation that does not have to increase owing to the frequency spectrum crossover is interior, mobile station speed equals 20 metre per second (m/s)s), less than the quantization error of measurement data itself.And, do not influence power control and switch comparison procedure because it is as one man less than normal.

Fig. 1 is the handover schematic diagram, Fig. 2 is the system function of consistent averaging method, Fig. 3 is the function of exponential average method, Fig. 4 is the probability of the quick saltus step of shadow fading, Fig. 5 is a map, Fig. 6 is that the different disposal method compares same sample, and wherein level value unit is dBm, and unification has numerically added 110.

Specify the present invention and effect thereof with simulation example below.The environment and the method for 1 simulation

For the confirmatory measurement processing method of data, we have carried out simulation study (class gsm system (GSMSpecification 05.08 version 5.4.0)).Adopt map shown in Figure 5, between two straight line paths of block two adjacent base stations are arranged respectively, be masked as base station 1, base station 2, base station 3 and base station 4.Synchronization has only single travelling carriage to move between two base stations, and the candidate target cells of switching can be uncertain.Be in the travelling carriage that B orders and both can switch to base station 2, also can switch to base station 3 or base station 4.In emulation, think that up-downgoing is a balance, only adopted the measurement data on one of them direction link, mobile channel is not considered the influence of rapid fading, only consider slow fading and path loss, and adopted COST 231 forecast models (GSM Specification 05.22 version 5.0.0).The processing of measurement data was to carry out in the cycle with 0.48 second.The sample performance of 2 Measurement and Data Processing

At first investigate the performance of different Measurement and Data Processing methods in the route sample that turns round.Adopted the processing method of four kinds of measurement data in the emulation.

(1) sliding window average (average length is 4)

(2) (average length is 4, α=0.5) asked on average in exponential weighting

(3) low-pass filtering (N=10, the filtering of second order Butterworth)

(4) low-pass filtering adds knowledge representation (N=10, the filtering of second order Butterworth, Δ=5)

Fig. 6 has compared the result of four kinds of Measurement and Data Processing methods to same random sample (travelling carriage moves along path A-B-C, leaves base station 1 gradually, does not do any switching, and speed is 20 metre per second (m/s)s, 200 meters after right-angle turning).As can be seen from the figure, the variation of primary signal intensity level also has bigger fluctuation when totally diminishing gradually, on average smoothly can realize preferably smoothly, but the quick variation of signal strength signal intensity is had bigger time delay (the corner time delay is roughly 1 second); Exponential average will descend to some extent to smoothness properties, but to the fast-changing time delay of signal strength signal intensity make moderate progress (the corner time delay is roughly 0.5 second); Low-pass filtering is when reaching than the better smoothness properties of sliding window averaging method, and time delay is also bigger, has only just realized good level and smooth and low time delay (considering that from the angle of switching its time delay is almost 0) in the method for filtering binding pattern identification simultaneously.The switching of 3 Measurement and Data Processing is inquired into

Investigate of the influence of Measurement and Data Processing method for handover.The trigger condition of switching is the incoming level big 5dBm of the incoming level of adjacent area in the one-shot measurement data processed result than this sub-district.Call drop adopts the Radio Link timeout mechanism of gsm system.The Radio Link timeout mechanism is: system calls out each and keeps a counter, initial value is its maximum, does not correctly receive a SACCH piece (transmission unit of the channel of channel associated signalling at a slow speed of gsm system, per 4 frames, one multi-frame each time, be 0.48 second multiframe period), counter subtracts 1; When counter during less than maximum, whenever correctly receive a SACCH piece, counter adds 2.When counter less than zero the time, system will think that this link is unavailable, and define call drop one time.Investigate the characteristic of Measurement and Data Processing method respectively with two kinds of methods in the different application occasion:

(1) the average signal strength level of adjusting desirable switching point (the B point among Fig. 6) is pact-95dBm,

Travelling carriage moves along a straight line 400 meters, investigates the smoothness properties of data processing method.

(2) the average signal strength level of the desirable switching point of adjustment is about-95dBm, and travelling carriage is transported along straight line

Right-angle turning after moving 200 meters, rectilinear motion is 200 meters again, investigates data processing method to turning round

The resolution ability of effect.Table 1 has provided the sliding window averaging method and low-pass filtering adds the corresponding statistics of Knowledge Representation Method.

The analog result of table 1 switch data preprocess method

	Rectilinear motion *		Right-angle turning **
					Sliding window is average	Low pass adds detection	Sliding window is average	Low pass adds detection
	Average switching times	1.73	?1.70	?1.15					?1.41
Average cutting off rate (%)					0.23	?0.29	?15.2	?5.1

^*The overtime initial value of Radio Link is made as 1, and motion velocity of mobile station is 2 metre per second (m/s)s

^*The overtime initial value of Radio Link is made as 10, and motion velocity of mobile station is that 20 metre per second (m/s)s can be seen from last table: (1) the smoothness properties simulation, the recognition methods of filtering binding pattern shows as average than sliding window really

The better smoothness properties of method.The decline of average switching times when it shows the travelling carriage rectilinear motion,

Though the amplitude that changes is fewer, because this change nearly all is by very little part conversation

Contribution, for these conversations, the rising of speech quality is considerable.(2) under the straight-line situation of travelling carriage, even in the overtime situation that is made as minimum value 1 of Radio Link

Down, average cutting off rate is not greatly improved.The overtime setting in practice of Radio Link is to want

Be far longer than 1, therefore since the raising of the level and smooth cutting off rate that causes to the raising of system's cutting off rate

Be negligible in practice.(3) in the simulation of the effect of turning round, the cutting off rate of filtering binding pattern recognition methods is average than sliding window

Method has descended greatly, and this shows that the quick saltus step of detection signal strength is to avoid very big one one

Divide the call drop that turns round and cause.Further corrective measure is the detection that improves neighboring area signal, the adjacent area

The Measurement and Data Processing and the Measurement and Data Processing branch of Serving cell come and do.Because the adjacent area does not have merit

Rate control problem can be strengthened the level and smooth of nearby region measurement data and to the detection of the effect of turning round, reduces

The probability of select target sub-district improperly in the switching after turning round is because select target improperly

The sub-district has constituted most call drops and unnecessary switching.(4) cutting off rate that effect causes of can also finding to turn round in the simulation of the effect of turning round is quite high really

, it is one of principal element that influences in the Microcell system's cutting off rate.

Claims (2)

1, the preprocess method of wireless measured data in a kind of cell mobile communication systems, it is characterized in that adopting fluctuation endless low pass filter elimination period of waves less than signal period N, wireless measured data is carried out smoothing processing, then the saltus step between continuous two data is detected, data after level and smooth still have the saltus step greater than signal strength signal intensity whereabouts saltus step threshold delta, then, directly recover received signal without smoothly.

2, the preprocess method of wireless measured data according to claim 1 is characterized in that the low pass filter that adopts is a second order Butterworth wave digital lowpass filter.

Images